A couple of times I've run into a situation where I have RLocked an RWMutex and then
need to Lock it to modify the thing it is protecting, and then want to go back to an
RLock to use it. A typical example is grabbing the RLock to examine cached data, then
noting it is stale and then needing to Lock to update it.
Unfortunately the obvious approach (RUnlock + Lock, then modify, then Unlock + RLock) is
racy because the condition that caused you to promote to a write lock may have changed
between the Unlock and the RLock. This necessitates a loop structure.
If there were Promote and Demote methods then that shuffle could be avoided. The
semantics are:
- Promote is equivalent to mu.Lock() with a concurrent mu.RUnlock(), where the Lock happens-before the RUnlock.
- Demote is equivalent to mu.RLock() with a concurrent mu.Unlock(), where the RLock happens-before the Unlock.
